Microwave (3.4GHz) LNA design with GaAs fet - impedance matching trouble(s)

[Yansi]

Hello,
its me again with another mind exercise* designing a microwave LNA.  As I have acquired a bunch of ATF-10136 GaAs fets from some old RF junk, I would like to attempt to make a LNA for the first two radioamateur bands, 1.296GHz and 3.4GHz, so I can increase the chooch factor of my microwave band receiver thing.

(* Because it will not likely end up with anything usable)

Here is the datasheet for the ATF-10136: https://www.qsl.net/n9zia/trans/ATF10136.pdf
Here is an interesting app note I have found about designing an S-band LNA with just this transistor: http://rfelektronik.se/manuals/Datasheets/ATF-10136-TR2.pdf

Please mind you, I have no formal knowledge/education in RF, it is just where I like to poke around more than I should. I have only basic understanding of what is happening there, so please bear with me.

I am now trying to replicate the input matching circuit for operation in the 3.4GHz band. So in stage one, I have plotted the available GammaOpt values and obtained the  GammaOpt values for 3.4GHz, which are 0.463 and 130 degree, according to the smoothly interpolated plot. The graph I have made is attached.  So the optimal reflection coefficient calls for 21.9ohm + 20ohm inductive  Which I have calculated using z = (1+Γ)/(1-Γ).

Following the application note, I have calculated I need a quarter wave TRL of 33ohm impedance to match the real part of 21.9ohms:
Zo = sqrt(Z1*Z2) = sqrt(50*21.9) = 33ohm
Okay, so far it sounds good to me.

Now I need to match the 20ohm inductive. That wants me to use a series inductor of 0.94nH (X_L=2πfL). Unfortunately and exactly as written in the appnote, this is difficult to make, so I followed the appnote further to replace the series inductor with a shunt capacitor:
Xc = Zo² / X_L = 33² / 20 = 54.5ohm = Xc

The capacitor needed is therefore 0.86pF (Xc = 1/2πfC) . Which is a pita because I do not know how this could be implemented in real circuit either.

What should I do in this case? Is there any other possibility to make  the impedance to mach? So far the 33ohm quarter wave TRL is pretty easily manufacturable, unfortunately I do not know what to do about the series inductor (0.94nH) or shunt capacitor (0.86pF). Both values are very small and I think very impractical for any kind of lumped element component .

Thank you for help
Y.

[David Hess]

The capacitor needed is therefore 0.86pF (Xc = 1/2πfC) . Which is a pita because I do not know how this could be implemented in real circuit either.

Wouldn't this be implemented as a microstrip structure?

Even at 1GHz, I have seen microstrip structures trimmed with an x-acto knife for impedance matching.

[Yansi]

Probably yes? But I am not the one who knows how to do those. 
Even the LNA in the appnote at the end of the document, has these made by some stubs. However the appnote does not discuss how those have been calculated. (As is usual with these microwave appnotes, there is just a bit of basics, then long way nothing and a result not matching the discussion from the beginning).

Looking at their amplifier,  I can't even find the 44ohm quarter wave resonator in the input.  44ohm should be visibly thicker than 50ohms. Don't tell me that its the thick bit right under the gate. No way in hell that is a quarter wave length on 2.3GHz.

I am just confused about how that whole structure was created.

[chrisl]

Looks like the layout was done based on the schematic shown in the figure 4.
They used the simulator to optimize the design so the initial matching network they talked about in the beginning no longer applies here.

[Yansi]

Well but then the overall appnote is useless, if they do not tell you how it was optimized.  Very unlikely you can suck the stub lengths and positions out of your finger, or to just "try them all"  to see which one fits.

[T3sl4co1l]

Right, usual design and testing procedure is to leave the stubs long and trim them to length (or leave them short and put some floating pads off the end so you can jumper to them as needed), and use trimmer caps and stretchable air core inductors where possible (which may not be very possible at this frequency, of course).  Go at it with a soldering iron, screwdriver, pliers and a VNA, until you've got the input and output match tuned as you need them.

Tim

[yl3akb]

Well but then the overall appnote is useless, if they do not tell you how it was optimized.  Very unlikely you can suck the stub lengths and positions out of your finger, or to just "try them all"  to see which one fits.

App note clearly states that initial design is obtained with help of Smith chart. For example attached picture shows, how shunt C=2.6pF and series line with Z0=44 and L=90deg can easily be obtained if we need to match Gamma=0.65@59deg at 2.3GHz.
 

[G0HZU]

Looking at their amplifier,  I can't even find the 44ohm quarter wave resonator in the input.

It's the long horizontal trace that leads into the gate. In other words, look at the PCB artwork text that says 2.3/2.4 and it's the horizontal trace just below this. It should then be easy to work out where the other transmission lines are. Note that the bias feed connections are not a direct connection as shown in the schematic. There is a skinny quarterwave line used here on both the input and the output.

Note also that this is a very old app note and the linear simulation tool they used was a very basic node based system. However, it looks like they did include models for the steps and junctions for the microstrip lines. So their simulation isn't quite as straightforward as using basic models that tie together the various 90deg matching sections, the stubs and and the skinny 90 degree bias feed sections. So their final simulation will have been tweaked after taking into account their enhanced models for the step changes in the microstrip and the microstrip junctions.

Ideally, you would want to use a more modern tool to simulate the layout such as Sonnet EM as this will manage all this stuff for you. This will be even more important up at 3.4GHz. Also, I would recommend you try playing with a modern PHEMT device such as the classic ATF54143 instead. The ATF 54143 is sadly going obsolete but you can still get them and the models are modern/proven and the device can be run from a single supply. Much easier all round I think.

The old s2p + noise model of the ATF10136 from HP is probably fine but there is some risk that it might not be as accurate up at 3.4GHz. You also need a dual power supply for this device. The other thing to watch out for is how you manage the source degeneration. This will have quite an impact on the gain and the input match and stability. Why not practise with the modern ATF54143 PHEMT and then come back to the ATF10136 when you have gained some experience from the 54143 part?